Instructions to use Akahsizrr/mini-deepseek-v4-flash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Akahsizrr/mini-deepseek-v4-flash with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Akahsizrr/mini-deepseek-v4-flash")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Akahsizrr/mini-deepseek-v4-flash", device_map="auto")

MLX

How to use Akahsizrr/mini-deepseek-v4-flash with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm
# if on a CUDA device, also pip install mlx[cuda]

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("Akahsizrr/mini-deepseek-v4-flash")

prompt = "Once upon a time in"
text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

vLLM

How to use Akahsizrr/mini-deepseek-v4-flash with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Akahsizrr/mini-deepseek-v4-flash"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Akahsizrr/mini-deepseek-v4-flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Akahsizrr/mini-deepseek-v4-flash

SGLang

How to use Akahsizrr/mini-deepseek-v4-flash with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Akahsizrr/mini-deepseek-v4-flash" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Akahsizrr/mini-deepseek-v4-flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Akahsizrr/mini-deepseek-v4-flash" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Akahsizrr/mini-deepseek-v4-flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

MLX LM

How to use Akahsizrr/mini-deepseek-v4-flash with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Generate some text
mlx_lm.generate --model "Akahsizrr/mini-deepseek-v4-flash" --prompt "Once upon a time"

Docker Model Runner
How to use Akahsizrr/mini-deepseek-v4-flash with Docker Model Runner:
```
docker model run hf.co/Akahsizrr/mini-deepseek-v4-flash
```

mini-deepseek-v4-flash / validation /tests_fuse2_mlx.py

Akahsizrr

Publish compatibility source validation/tests_fuse2_mlx.py

f76fbf8 verified about 17 hours ago

Raw

History Blame Contribute Delete

1.04 kB

	from __future__ import annotations

	import mlx.core as mx

	from microscope.fuse2_mlx import Model, ModelArgs


	def main():
	args = ModelArgs(
	model_type="fuse2",
	hidden_size=32,
	num_hidden_layers=2,
	intermediate_size=64,
	num_attention_heads=4,
	rms_norm_eps=1e-6,
	vocab_size=128,
	num_key_value_heads=2,
	max_position_embeddings=64,
	rope_theta=1000000.0,
	head_dim=8,
	tie_word_embeddings=True,
	experts_per_layer={"0": [0, 1]},
	expert_hidden_size=32,
	expert_intermediate_size=16,
	)
	model = Model(args)
	ids = mx.array([[1, 2, 3]])
	uncached = model(ids)
	cache = model.make_cache()
	model(ids[:, :-1], cache)
	cached = model(ids[:, -1:], cache)
	mx.eval(uncached, cached)
	error = float(mx.max(mx.abs(uncached[:, -1] - cached[:, -1])))
	if error > 1e-4:
	raise AssertionError(error)
	print({"status": "validated", "max_logit_error": error})


	if __name__ == "__main__":
	main()